AI Briefing: With OpenAI in the spotlight, others release their own new AI models

A day later, another new large language model was released by Inflection AI, the AI startup behind a popular chatbot called Pi. The company — co-founded by Mustafa Suleyman and LinkedIn co-founder Reid Hoffman — claims its new Inflection-2 model beats Google’s PaLM in a number of key benchmarks. However, the company says it still needs to undergo various steps before it begins powering Pi in order to make sure it’s a “helpful and safe personal AI.” (Suleyman is also the cofounder of Deepmind, an AI startup Google acquired in 2014.)

Beyond the startups, Microsoft also announced its own updated LLMs. The tech giant debuted Orca 2, which comes in two sizes that are both much smaller than OpenAI’s GPT-4 and Meta’s LLAMA-2. In a blog post, Microsoft researchers said the smaller model was capable in ways previously only seen in larger models and in some cases in even outperformed them.

“A key insight behind Orca 2 is that different tasks could benefit from different solution strategies (e.g. such as step-by-step processing, recall then generate, recall-reason-generate, extract-generate, and direct answer) and that the solution strategy employed by a large model may not be the best choice for a smaller one,” the researchers wrote. “For example, while an extremely capable model like GPT-4 can answer complex tasks directly, a smaller model may benefit from breaking the task into steps.”

AI news:

• Last week’s chaos within OpenAI raised a range of new questions about how to approach generative AI. The startup’s mercurial state has some companies reexamining their approaches to adopting large language models. Some companies are considering diversifying their LLM providers to avoid being overdependent on a single company, which could also help other companies catch up. Other observers said the strong reaction to Sam Altman’s ouster — and his comeback — showed the role of people and emotions when it comes to adopting new tech. The leadership changes were also seen by some as a reason for holding foundation model providers more accountable through new AI regulations.

• OpenAI and Microsoft are facing yet another lawsuit related to generative AI. The proposed class action lawsuit was filed on behalf of Julian Sancton, a journalist at The Hollywood Reporter, who alleges OpenAI violated copyright laws by scanning a book he authored and using it to train its GPT models.

• The Federal Trade Commission voted to approve the use of a legal action similar to a subpoena, which the agency will be able to use when investigating AI systems. According to the agency, the tool — known as a civil investigative demand (CID) — will help the FTC “obtain documents, information and testimony.”

Amazon’s plan for an “AI ready” workforce

Ahead of Amazon’s re:Invent cloud computing conference this week in Las Vegas, the company debuted a range of new educational initiatives to help 2 million adults and students become “AI ready” by 2025.

Along with eight new free AI courses for a mix of technical and non-technical audiences, Amazon also announced a new collaboration with Code.org that includes a “code dance party” for kids to learn to code virtual music videos. Another new initiative aims to teach generative AI to more than 50,000 high school and college students through a new partnership with the online learning platform Udacity.

According to Amazon’s new study about how AI will transform the workplace, 73% of respondents mentioned hiring AI-skilled workers is a priority, but only a fourth said they’ve been able to find the talent they need. (Last week, LinkedIn also released its own separate report about AI and the future of work.)

The pace of AI innovation is also leading Amazon to develop and deploy AI courses faster than it has for other types of tech. Jenni Troutman, director of Products and Services for AWS’s Training and Certification, said the goal is to make courses as broadly available as possible and to provide ways for people to learn at their own pace. To make courses relevant to the pace of change, she said Amazon is focusing on foundational information that will still apply even as AI evolves. “Everybody knows it’s going to transform the way we work, but nobody knows for sure how other than the things we’re already doing,” Troutman said.

There seems to be plenty of interest among workers in learning new AI skills. Amazon’s survey found 80% of workers — including two-thirds of workers over 55 — expressed interest in acquiring new AI skills. There also could be added incentive for workers to spend time learning new AI skills — Amazon’s survey found employers were willing to pay an average of 47% more for IT workers with AI talent. Other industries surveyed also expressed a willingness to pay more for AI knowledge. Sales and marketing companies surveyed said they’d be wiling to pay 43% while business operations firms mentioned 41%.

According to Troutman, customers consistently ask questions about generative AI related to how it works, what’s possible and what it means for business strategy. However, she said it can be challenging for people to figure out how to apply AI effectively and responsibly if they don’t first understand how it works.

“The idea of jumping in and trying to use something before really understanding what it is can be challenging here,” Troutman said. “What I would encourage people to do is go out there and figure out as much as you can about what AI, ML and GenAI is and how it can how it can be used before jumping in to actually use a tool. Then you can quickly move to leveraging it.”

Other AI stories from across Digiday:

• Earlier this month, the influencer agency Billion Dollar Boy created a new division called Muse that integrates generative AI and other tech into various offerings.

• A new video explainer details how AI is creating copyright concerns and how companies, regulators and creators are looking to deal with the challenges.

• Lizzie Widhelm, svp and head of ad innovation and B2B marketing at SiriusXM Media, spoke with Digiday for a podcast episode about selling audio ads programmatically, the future of improving audience targeting and how generative AI might help automate programmatic audio ad sales.